MP352notes IndexNotation A 01

p.
1 Tensor notation or index notation
1 These notes
Here we introduce tensor notation or relativistic index notation.
We probably previously alluded to this notation multiple times al-
ready, e.g., referring to the components of 4-vector A with super-
scripted (upstairs) indices, like Aµ , and referring to inner products
as Aµ Bµ or Aµ Bµ . What does it mean to have subscripted (down-
stairs) indices? We will define this soon.
The index notation we are introducing — involving superscripts nad
subscripts — is also known as tensor notation. We will be introduc-
ing tensors as well in this chapter.
2 Latin & Greek, Einstein notation

An event or a spacetime point is a 4-vector
X µ = (ct, x, y, z) with µ = 0, 1, 2, 3;
so that X 0 = ct, X 1 = x, X 2 = y, X 3 = z.
We have previously complained that the symbol X µ is often used to

represent the complete 4-vector, not just the µ-th component. (There
is nothing we can do about it, except for getting used to the indignity
of slightly inconsistent notation.)
Similarly for the 4-momentum
Pµ = ( E/c, p x , py , pz ) with µ = 0, 1, 2, 3;
so that P0 = E/c, P1 = p x , P2 = py , P3 = pz .
Of course, there are texts that place time as the 4th coordinate, so that µ
runs from 1 through 4, instead of running from 0 through 3 like we have. The
4-momentum components in such a convention would be
E
P1 = p x , P2 = p y , P3 = p z , P4 = .
c
For the Minkowski indices, we usually use lower-case Greek letters,

e.g., µ above. Sometimes, in the same calculation you might have to
introduce indices that run only over the spatial part, i.e., only from 1
to 3. It is a common convention to use lower-case Latin letters (i, j,
k,. . . ), i.e., lower-case letters from the English-language alphabet, to
mean indices that run over the three spatial components, excluding
the temporal component.
p. 2 Tensor notation or index notation
We are familiar with writing Lorentz transformations (from inertial

frame Σ to inertial frame Σ̃) as the matrix equation
   
ct̃ ct
 x̃  x
  = Λ 
 ỹ  y
z̃ z
for spacetime coordinates. Here Λ is a 4 × 4 transformation matrix.

In the index notation that we are developing, this is written as
X̃ µ = ∑ Λµ ν X ν .
ν
You recognize this as being just the definition of matrix multiplica-

tion, with Λµ ν representing the element in the µ-th row and ν-th
column of of Λ. The second index is however written downstairs,
for reasons that will become clear later on. In the expression on
the right, the summation index is ν. A summation index is called
a ”dummy index” because it can be replaced by any other symbol
without changing the meaning of the equation.
All these indices are going to get painful. Let’s make our lives easier
with the notational trick Lorentz
trans-
µ
X̃ = ∑Λ µ
νX
ν
= Λ νX .
µ ν
(1) forma-
tion, in
ν
tensor
We just omitted the summation symbol! The idea is that, when- nota-
ever we see an index repeated, we will understand that this index tion
is summed over. This is called the Einstein summation convention.
It will save us writing summation symbols with indices under them,
but we will have to remember to sum whenever there is an index
appearing twice. E.g., the 0-th component of X̃ is
X̃ 0 = Λ0 ν X ν = Λ0 α X α = Λ0 0 X 0 + Λ0 1 X 1 + Λ0 2 X 2 + Λ0 3 X 3 .
Since any repeated index is a summation index and hence a dummy

index, it doesn’t matter whether we call it ν or α!
In our example, the index that is repeated appears once upstairs and once
downstairs. This will be true in general: an upstairs index is contracted with a
downstairs index.
Summing over the repeated index is often called contracting the two indices.
The summation convention is widely used also in situations where no

distinction is made between upstairs and downstairs indices. For example, in
discussing non-relativistic physics in Euclidean space, you might see the
expression Mij A j to mean ∑ j Mij A j , i.e., a matrix-vector multiplication. You
might see this, e.g., in an advanced text on non-relativistic classical

mechanics, with i, j running over the three spatial directions, 1, 2, 3.
I read somewhere that Einstein regarded the summation convention to be one

of his greatest contributions.
By definition, any 4-vector has to transform the same way under

LT’s. Thus if A is any 4-vector, it transforms as Ãµ = Λµ ν Aν . From
Eq. (1) we note, by differentiating with respect to X ν on both sides,
that
∂ X̃ µ
Λµ ν =
∂X ν
so that the transformation equation for a 4-vector with upstairs-index
is
∂ X̃ µ ν
Ãµ = Λµ ν Aν = A .
∂X ν
3 Indices upstairs and downstairs — contravari-

ant and covariant
It’s time to introduce 4-vectors with downstairs indices. The 4-vectors
we have used till now, with upstairs indices, are called contravariant
vectors, and the superscript indices are contravariant indices. Asso-
ciated with each contravariant 4-vector is a dual 4-vector which we
call a covariant 4-vector, written with an index downstairs. The sub-
script indices are covariant indices.
Instead of writing contravariant vectors and covariant vectors, we
might sometimes just call them vectors and covectors, respectively.
3.1 Raising and lowering

The metric tensor is written either as gµν or as gµν . Given a con-
travariant 4-vector Aµ , the corresponding covariant 4-vector is de-
fined as
Aα = gαβ A β
where of course the summaton convention is understood. The co-
ng
variant 4-vector is obtained by contracting the contravariant index

eri
low
with one of the indices of the metric tensor. This is colloquially called
“lowering the index.”
You can see that the components of the covariant vector Aµ are closely
related to the components of the corresponding contravariant vector
Aµ . In our convention, the diagonal elements of gµν are (1, −1, −1, −1);
therefore
A0 = A0 , A1 = − A1 , A2 = − A2 , A3 = − A3 .
The timelike component is unchanged by index lowering, while the

spacelike components get their signs rudely flipped. Using the con-
vention that Latin indices represent the spatial components, we could
write this as A0 = A0 , Ai = − Ai .
For example, the elements of the covariant vector corresponding to a
4-momentum are P0 = E/c, P1 = − p x , P2 = − py , P3 = − pz .
In the other convention for the metric tensor (positive-trace conven-
tion), the temporal component would have its sign flipped while the
spatial components would remain unchanged. (A0 = − A0 , Ai = Ai )
The covariant vector was obtained from the contravariant vector through
multiplication by the matrix gµν . One should be able to obtain the
contravariant vector from the corresponding covariant vector through
multiplication by the inverse of gµν . Fortunately, the metric tensor is
its own inverse. Thus we have
Aρ = gρλ Aλ .
ing
s
rai
We have “raised” an index using the metric tensor.
Note that the inverse of gµν is written with indices upstairs. This
guarantees that an upstairs index is contracted with a downstairs
index.
3.2 Inner products and norms

The inner product of 4-vectors A and B can be written as
A0 B0 − A1 B1 − A2 B2 − A3 B3 = A0 B0 + A1 B1 + A2 B2 + A3 B3 = A µ B µ .
The summation convention has been used in the last step. The in-
ner product is thus written in a remarkably simple form, using the
notation we’ve introduced.
It is easy to show ( please do ) that the following forms are equiva-
lent, all representing the same inner product:
Aµ Bµ = Aµ Bµ = gµν Aµ Bν = gµν Aν Bµ = gµν Aµ Bν = gαβ A β Bα .
Hopefully, it should be obvious that Aµ Bµ = Aα Bα = A β B β =

Aν Bν = Aλ Bλ . A repeated index is summed over and hence is a
dummy index — in principle any symbol can be used for it. But
please restrict to smaller-case Greek symbols — this is conventional
for Minkowski indices running from 0 to 3.
Similarly, you should be able to show ( please do! ) that the norm of
the 4-vector A is
Aµ Aµ = Aµ Aµ = gµν Aµ Aν = gµν Aµ Aν .
3.3 Transformation
We have defined covariant vectors through the index-lowering op-
eration. It is also common to define covariant vectors through their
transformation properties.
We have seen that a contravariant vector Aµ transforms under Lorentz
transformations as
∂ X̃ µ ν
Ã = Λ ν A =
µ µ ν
A .
∂X ν
The corresponding covariant vector transforms as Transformation
of co-
ν ∂X ν variant
Ãµ = Λ−1 Aν = Aν (2) vectors
µ ∂ X̃ µ
Covariant vectors transform oppositely to how contravariant vectors
transform.
We can show this transformation rule, taking Aσ = gσβ A β to be the
definition of the covariant vector. It’s a bit clumsy and is worked out
in the box below.
Since Aµ = gµν Aν , in the frame Σ̃ we have
Ãµ = g̃µν Ãν = gµν Ãν = gµν Λν σ Aσ = gµν Λν σ gσβ A β
Looks like the transformation matrix is gµν Λν σ gσβ . Remembering the

summation convention, this is the product of three matrices, g, Λ and g−1 .
Symbolically
β
Ãµ = gΛg−1 Aβ
µ
The indices ν and σ are summed over and so don’t appear any more, while
the µ and β indices are written with their order unchanged, which makes the
covariant index of the transformation matrix the first (‘row’) index, and the
contravariant index the second (‘column’) index. To maintain our convention,
let’s take the transpose, so that the contravariant index of the transformation
matrix becomes the first index:
T β
−1
Ãµ = gΛg Aβ
µ
Now from the definition of the Lorentz transformation:

−1 T
ΛT gΛ = g =⇒ gΛg−1 = ΛT = Λ −1
T
so that gΛg−1 = Λ −1 .
β
Thus the transformation matrix is Λ−1 , so that Ãµ = Λ−1 µ
Aβ .
We have used ‘index lowering’ to define covariant vectors, and from

that proved that covectors transform ‘oppositely’ as vectors. Often,
however, Eq. (2) is taken as the definition of covectors, and the low-
ering/raising operations are regarded as a notational trick.
To summarize, the transformation of vectors and covectors under a
Lorentz transformation Λ are
β
−1
Ã = Λ
µ µ
νA
ν
, Ãµ = Λ Aβ (3)
µ
The transformation equation for covectors is sometimes written as

β
Ãµ = Λµ β A β , where it is understood that Λ−1 µ = Λµ β . This
notation relies on taking care of the horizontal placement of indices
in addition to their vertical placement. I find this clumsy, especially
as one might reasonably decide to interpret Λµ β as the transpose of
Λ β µ . In (3), we have used the contravariant (upper) index as the first
(row) index in both transformations.
To avoid worrying about horizontal placement of indices altogether,
it’s convenient to remember instead the following forms of the trans-
formations:
∂ X̃ µ ν ∂X β
Ãµ = A , Ã µ = Aβ (4)
∂X ν ∂ X̃ µ
This form highlights that a vector (contravariant vector) transforms
the same way as spacetime coordinates (spacetime 4-vectors). This
is of course how we defined vectors. It also shows that covectors
transforms oppositely to (contravariant) spacetime 4-vectors.
Contra and Co — what’s with these names?
It’s amusing that regular vectors are ‘contra’ and the artificial new objects
we’ve introduced are ‘co’, which sounds more positive than ‘contra’.
The names contravariant and covariant relate to transformation properties. A
coordinate frame can be described by unit vectors, or basis vectors. It turns
out that contravariant vectors transform oppositely to the transformation of
unit vectors, hence ‘contra’. Covariant vectors transform the same way as the
unit vectors, hence ‘co’. In describing flat Minkowski space, unit vectors are
not very necessary, and so will not introduce them.
3.4 The metric tensor and its inverse, notation

We have noted that gµν is the inverse of gµν . It is denoted by g instead
of g−1 because the two happen to be the same in special relativity in
Cartesian coordinates.
Even when they are not the same (e.g., in general relativity, or if using
curvilinear coordinates), using g−1 would be unnecessary because
the indices appearing upstairs tell us that gµν is the inverse.
Treating them as matrices, multiplying the two should give the 4 × 4
identity matrix:
µ
gµβ g βν = δν
where the right side is the Kronecker delta, = 1 when µ = ν and = 0
when µ 6= ν. In the present context, the Kornecker delta has to be
written with one index up and one index down, but the horizontal
placement of the indices doesn’t matter as it is a symmetric matrix.
gµν versus ηµν
It is common to make a notation distinction between the metric tensor of
non-flat spacetime (as in general relativity), and the metric tensor in flat
spacetime (as in special relativity). In the former case, the metric itself
depends on spacetime coordinates, while in the latter, we have the constant
metric we are used to in this semester. It is common to use gµν for the
general case and ηµν for the special (flat) case; so you will find ηµν in many
texts. For our purposes, it does not matter.
3.5 Maybe think of covectors as row vectors?

(Optional subsection)
We have introduced of a second type of index and a second type
of vector. We have already seen a gain in conciseness due to this
notation — the inner product of A and B being written as Aµ Bµ .
A possibly useful way to think about this is to think about contravari-
ant vectors as column vectors (4 × 1 matrices) and covariant vectors
as row vectors (1 × 4 matrices). Lower indices are column indices
and upper indices are row indices. The contraction (implied sum-
mation) of the index in the expression Aµ Bµ can be regarded as a
multiplication between a row vector and a column vector to yield a
1 × 1 number:
 0
B
 B1  0 1 2 3
A0 A1 A2 A3   B2  = A0 B + A1 B + A2 B + A3 B .

B3
The Lorentz transformation equation Ãµ =  0   0

Ã A
Λ ν A can then be regarded as a matrix-vector  Ã1 
µ ν
 A1 
multiplication — a square matrix multiplies a col-   Ã2  = Λ  A2 
  
umn vector to give a column vector. The trans- Ã3 A3
formation matrix Λ ν has an upper and a lower
µ
index, serving as a row index and a column in-

dex respectively. The contraction is between the column index of Λ
and the row index of the contravariant vector it is mutliplying — this
matches the usual rule of matrix multiplication.
This is satisfying. What about the transformation rule for covectors:
β
Ãµ = Λ−1 µ A β ? Now, on the right, the row index of Λ contracts
with the column index of the covector. This is the standard multipli-
cation of a row vector from the left with a matrix, made more visible
by re-ordering the right side:
β β
Ãµ = A β Λ −1
= ∑ Aβ Λ −1
.
µ µ
β
In the row-vector picture, this can be visualized as

 
 
Λ −1
 
Ã0 Ã1 Ã2 Ã3 = A0 A1 A2 A3 

.

 
So far this picture has been quite satisfying. While the analogy is
useful to some extent, it has its limits, which we next illustrate.
Let’s look at the lowering operation — We obtained covariant vectors
from contravariant vectors using the metric tensor to lower an index:
A β = g βµ Aµ . This equation cannot be regarded as a matrix-vector
multiplication. Since both indices of g are down, it would seem to
have two column indices, which would be a strange way to describe
a matrix. The action of index lowering is to transpose the column
vector and then flip the signs of the spatial part:
 T  
A0 1 0 0 0
 × 0 −1 0
 A1   0 
Ã0 Ã1 Ã2 Ã3 = 
 A2  0 0 −1 0 

A3 0 0 0 −1
or equivalently, flipping signs of the spatial components first and

then taking the transpose. Either way, the lowering operation is dif-
ficult to write as a simple matrix multiplication. This illustrates that
the row-vector picture is rather limited. The index notation is actu-

ally richer and more powerful than the matrix picture, and allows us
to express relationships that don’t have a nice interpretation in the
row-vector/column-vector language.
Elsewhere (outside Minkowski space or relativity), you might also meet this
concept of introducing duals to each vector in order to define inner products.
For example, consider the Euclidean dot product A ~ · ~B between 3-vectors A
~
~
and B. If you want, you could think of this as a matrix product between the
A written as a row vector and B written as a column vector. We could even
call the row-vector a covector. There is little advantage to introducing this
notation, so it’s not usually done.
In quantum mechanics, every state is a vector in Hilbert space, which in Dirac
notation we think of as the space of kets. A dual space is introduced:
corresponding to every ket |φ i is a bra h φ|. Thinking of kets as column
vectors and bras as row vectors, we can think of inner products in quantum
mechanics as arising from multiplying a row-vector and a column-vector:
 
φ1
φ  D
∗  2 =

hψφi = ψ1∗ ψ2∗ . . . ψD ∑ ψ∗ φα

 .. 
 .  α =1 α
φD
In quantum mechanics, this is a useful and widely accepted way to think

about ket-vectors and bra-vectors, as long as the Hilbert-space dimension D is
finite. When D is infinite (common in quantum mechanics!), you have to go
back to the abstract definition and the column-row picture of kets and bras
doesn’t work any more.
In discussing relativity and the Minkowski inner product, the column-row
picture of contravariant and covariant vectors is not as common, and, beyond
very simple manipulations, has limited use. We will soon be doing very funky
things with indices. The matrix notation won’t be able to keep up.
3.6 The 4-gradient and 4-divergence

In (Euclidean) vector calculus, we use the nabla operator or gradient
operator,
~ ≡ ∂ ∂ ∂
∇ , , ≡ ∂ x , ∂y , ∂z
∂x ∂y ∂z
to define the operations of gradient, divergence, and curl. If f is a
scalar field, then the 3-vector ∇ ~ f is the gradient of f . If V
~ is a vector
field, then the scalar field ∇~ ·V
~ is the divergence of V.
~ One can check
~ f transforms like a vector under
that, if f is a scalar under rotations, ∇
rotations, and that if V ~ transforms like a vector under rotations, then
~ ~
∇ · V is invariant under rotations.
We can generalize to the Minkowski world. The 4-derivative opera-

tor is
∂
∂µ = .
∂X µ
Notice that we have indexed ∂ downstairs, implying that it is a co-
variant vector operator. We will justify this shortly.
Given a Lorentz scalar, φ, its 4-gradient is the 4-vector

∂φ ∂φ ∂φ ∂φ
∂µ φ ≡ , , ,
∂X 0 ∂X 1 ∂X 2 ∂X 3
The subscript indicates that ∂µ φ is a covector rather than a vector.
Why? Consider a Lorentz transformation taking spacetime coordi-
nates X µ to X̃ µ . Under this transformation, the 4-gradient is trans-
∂φ
formed to , as φ is invariant. To express this quantity in terms
∂ X̃ µ
of derivatives with respect to the original coordinates X µ , we use the
chain rule for differentiation of implicit functions:
∂φ ∂X ρ ∂φ ∂X ρ ∂φ ρ
µφ =
∂g
∂ X̃ µ
= ∑ ∂X̃ µ ∂X ρ ∂X̃ µ ∂X ρ
= = Λ −1
µ
∂ρ φ.
ρ
Thus, ∂µ φ transforms as a covector, justifying our use of subscripts

on the ∂µ operator when differentiating with contravariant spacetime
coordinates X µ .
If one differentiates with respect to the components of the covariant
vector Xµ , one can obtain a contravariant gradient operator:
∂
∂µ ≡ = gµβ ∂ β .
∂Xµ
Applying the 4-gradient operator to a 4-vector Mµ = ( M0 , M ~ ) and
contracting the index gives a Lorentz scalar, which is the 4-divergence
of the vector:
∂ α M α = ∂0 M 0 + ∂1 M 1 + ∂2 M 2 + ∂3 M 3
∂M0 ∂M1 ∂M2 ∂M3
= + + +
∂x0 ∂x1 ∂x2 ∂x3
1 ∂M0 ~ ~
= +∇· M.
c ∂t
4-divergences are a convenient way to represent conservation laws
through continuity equations. If 1c M0 represents the density of some
quantity and the M ~ represents the corresponding current density,
then the continuity equation for this quantity would be expressed as
∂α Mα = 0. For example, the conservation of electromagnetic charge,
of mass, of electromagnetic field energy, etc. could be expressed in
this way.
4 Tensors in non-relativistic mechanics (3-tensors)

In 3D Euclidean physics, a scalar is a single-component object which
does not require indices (has zero indices); one could say that it has
1 = 30 elements. This is also known as a rank-0 tensor.
A (3-)vector has 3 = 31 elements and has 1 index. A vector is a rank-1
tensor.
Generalizing, a rank-2 tensor has 9 = 32 elements. It needs 2 indices.
If T is a rank-2 tensor, we could label it’s elements as Tij . It might
feel natural to pack the 9 elements into a 3 × 3 matrix. However, a
rank-2 tensor is more than a matrix — it is an object that transforms
in a certain way.
Under rotation of the coordinate frame, described by rotation matrix
R, a scalar (rank-0 tensor) remains invariant — the rotation matrix
R does not enter into the transformation relation. If f is a scalar, it
transforms as f 0 = f .
A vector (rank-1 tensor), transforms with a factor of R, e.g., if ~a =
( a1 , a2 .a3 ) is a vector, it transforms as
~a0 = R~a, or ai0 = Rij a j .
We’ve used the summation convention, even though we are doing
Euclidean physics.
Before writing down the transformation for a rank-2 tensor, let’s con-
struct an example. If ~a = ( a1 , a2 .a3 ) and ~b = (b1 , b2 .b3 ) are vectors,
then the combination of nine numbers
Tij = ai b j
forms a tensor. Not every rank-2 tensor can be decomposed in this
way into two vectors, but any two vectors can be combined (via such
an outer product) to construct a rank-2 tensor.
For the example above we can construct the transformation rule:
Tij0 = ai0 b0j = ∑ Rik ak ∑ R jm bm = ∑ Rik (ak bm )R jm = ∑ Rik Tkm (RT )mj
k m km km
0
=⇒ T = R T RT
We can write this more concisely using the Einstein summation con-
vention. Of course, then it becomes our responsibility to remember
what summations are implied:
Tij0 = ai0 b0j = Rik ak ∑ R jm bm = Rik ( ak bm )R jm = Rik Tkm (RT )mj

m
=⇒ T 0 = R T RT
We have found the transformation rule T 0 = R T RT for the special

case of rank-2 tensors that are outer products of two vectors. The
same transformation rule, using 2 factors of the transformation ma-
trix, holds for any rank-2 tensor, even those which cannot be decom-
posed into an outer product of two rank-1 vectors.
A tensor of rank higher than 2 is defined analogously. A rank-k ten-
sor has 3k elements and its transformation involves k factors of the
transformation matrix. For k > 2, writing the transformation as ma-
trix multiplication becomes clumsy or impossible, but it is easy to
write as a summation of appropriate indices. E.g., if B is a rank-4
tensor, then its elements transforms as
Bijkl = Rim R jn Rkp Rlq Bmnpq
with summation over repeated indices implied.
4.1 Examples of non-relativistic rank-2 tensors

The best-known example is probably the moment of intertia tensor
of an extended object with mass density ρ(~r ):
Z h i
3 2
Iij = d rρ(~r ) r δij − ri r j .
For a rigid body with extended structure, it relates the angular veloc-
~ to the angular momentum:
ity ω
Li = ∑ Iij ω j (5)
j
which could also be written as a matrix-vector multiplication.

This may be regarded as the analog of ~p = m~v, for angular motion.
When the object is complicated enough, the naive ~L = I ω ~ may not
hold, and the more general relation (5) is required. Note that con-
tracting one of the tensor indices with a vector index provides a vec-
tor.
Another non-relativistic example is the conductivity tensor. For sim-
ple materials, the current density ~J created by an applied electric field
~E is usually just proportional: ~J = σ~E. However, there are situations
(materials or geometries) when this is no longer true, and an electric
field can create currents in both longitudinal and transverse direc-
tions. Then we would write
Jk = ∑ σkl El .
l
Tensors also appear in the study of elastic properties of matter (the
elastic stress tensor or Cauchy tensor) and in the study of fluids (the
viscous stress tensor).
5 4-tensors and their transformations

In Minkowski space, a scalar has 1 = 40 element and is invariant
under Lorentz transformations. This is a rank-0 tensor. A vector
has 4 = 41 elements and its transformation involves 1 factor of the
transformation matrix ( M̃α = Λα β M β ). This is a rank-1 tensor.
Generalizing, a rank-2 tensor is a 2-index object with 16 = 42 ele-
ments. If T has two upstairs (contravariant) indices, then it trans-
forms as
T̃ αβ = Λα µ Λ β ν T µν
which is a straightforward generalization of the Euclidean case. Such
an object is obtained when we take an outer product of two 4-vectors
(tow rank-1 tensors): e.g., the 16 numbers Aµ Bν will transform as a
rank-2 tensor. However, given a rank-2 tensor, it cannot necessarily
be decomposed into an outer product of two 4-vectors.
Tensor indices can also be downstairs (covariant), and these indices
can be lowered or raised using the metric tensor:
gαµ T αβ = Tµ β , g βν T αβ = T α ν , gαµ g βν T αβ = Tµν .

If all indices are upstairs (downstairs), a tensor is contravariant (co-
variant); if some are upstairs and some downstairs, the tensor is of
mixed type.
Covariant components of tensors transform like covariant vectors,
i.e., with a factor of Λ−1 instead of a factor of Λ. Thus the mixed
tensor T α ν will transform as
ν
T̃ α β = Λα µ Λ−1 Tµ ν
β
and the totally covariant tensor Tρσ transforms as

µ ν
−1 −1
T̃αβ = Λ Λ Tµν
α β
You might find the Λ−1 ’s clumsy to write, especially to remember

the horizontal positioning of the indices. Maybe the following forms
are nicer:
∂ X̃ α ∂ X̃ β µν ∂ X̃ α ∂X ν µ
T̃ αβ = T , T̃ α β = T ν,
∂X µ ∂X µ ∂X µ ∂ X̃ β
∂X µ ∂X ν
T̃αβ = Tµν .
∂ X̃ α ∂ X̃ β
The concept of tensors, with contravariant and covariant indices, is
readily generalized to arbitrary rank. For example, the outer prod-
uct Aαβ Bµ is a rank-3 tensor with all contravariant indices: Aαβ Bµ =
Mαβµ . The outer product Aαβ Bµ ν is a rank-4 tensor with 3 contravari-

ant indices and one covariant index: Aαβ Bµ = N αβµ ν . However, if
you see the expression Aαβ Bµ β , you will know that the β index is
contracted (summed over), so that the result is a rank-2 contravariant
tensorr: Aαβ Bµ = Pαµ . If you see the expression Aαβ B β λ , someone
probably made a mistake, because a contravariant index is usually
not contracted with another contravariant index.
A rank-r contravariant tensor transforms as
∂ X̃ α1 ∂ X̃ α2 ∂ X̃ αr µ1 µ2 ...µr
T̃ α1 α2 ...αr = . . . T ,
∂X µ1 ∂X µ2 ∂X µr
or as
T̃ α1 α2 ...αr = Λα1 µ1 Λα2 µ2 . . . Λαr µr T µ1 µ2 ...µr .
A rank-r covariant tensor transforms as
∂X ν1 ∂X ν2 ∂X νr
T̃β1 β2 ...βr = . . . Tν ν ...ν ,
∂ X̃ β1 ∂ X̃ β2 ∂ X̃ βr 1 2 r
or as
σ1 σ2 σr
T̃β1 β2 ...βr = Λ−1 Λ −1 . . . Λ −1 Tσ1 σ2 ...σr .
β1 β2 βr
A tensor T µ1 µ2 ...µr β1 β2 ...β s of rank r + s, with r contravariant indices

and s covariant indices, will transform with r factors of Λ and s fac-
tors of Λ−1 . Not pretty.
6 Why is tensor notation useful?

Tensor notation leads to concise expressions in electromagnetism and
in (relativistic) quantum field theory. For example, in electromag-
netism, two of Maxwell’s equations become simply ∂µ F µν = J ν . How-
ever, tensor notation has a deeper benefit than conciseness.
If a physical law is expressed as an equality between Minkowski ten-
sors, it has the same form in all inertial frames.
For example, let us imagine that the equation or physical law
Pαβ γ = Qαβ γ + Rαβ γ + Aαβ Bγ
holds in frame Σ. Of course, for the equation to make sense, each

term needs to have the same number of un-contracted contravariant
indices and un-contracted covariant indices, and they have to be the
same indices. (In other words, only tensors of the same ‘size’ can
be meaningfully equal or added to each other, and tensor equality
implies element-by-element equality.) Do check that this is true in

the above made-up example.
Then in frame Σ̃, we will observe
P̃αβ γ = Q̃αβ γ + R̃αβ γ + Ãαβ B̃γ .
The reason is that, when transforming tensors, each uncontracted in-

dex brings in one factor of Λ or Λ−1 . Thus the same number of fac-
tors of Λ and Λ−1 appear with each term, and hence get cancelled.
The equation in the tilde-d frame ends up having the same form as
the equation in the no-tilde frame.
Thus, physical laws can be shown to be frame-invariant by express-
ing them in tensor form.
Often physicists say that an equation has been expressed in “Lorentz-
covariant” form, if it has been written using tensor form as above.
You might also hear “explicitly covariant” or “manifestly Lorentz-
covariant”, meaning that an equation or law is written in terms of
Minkowski tensors and hence is automatically known to have the
same form in all inertial frames. (This usage of the word covariant
is completely different from the same word used to describe down-
stairs indices. I am deeply sorry, but this double-usage is not my
fault.)
7 The Lorentz transformation in index nota-

tion
We have previously defined Lorentz transformations as those satis-
fying the matrix equation
ΛT gΛ = g (6)
where we have used matrix notation: The LT’s Λ are represented

here as 4 × 4 matrices, which act on event coordinates (or other 4-
vectors) represented as 4 × 1 column vectors. Applying Λ on the
spacetime coordinates of an event, x, gives the coordinates of the
same event as observed from another inertial frame: x̃ = Λx.
The relation (6) is derived from the condition that the quantity xT gx
is invariant under LT’s. The quantity xT gx is the Minkowski norm of
x, or alternately, the interval c2 t2 − |~r |2 from the origin at zero time.
Recall that the invariance of this interval follows from the invariance
of the speed of light.
We can do the same derivation now in index/tensor notation instead
of matrix notation:
If the 4-coordinate x µ transforms to x̃ α under a LT, then its norm

transforms to

x̃µ x̃ µ = gµν x̃ µ x̃ ν = gµν (Λµ α x α ) Λν β x β = gµν Λµ α Λν β x α x β
Since the norm has to be invariant, we need this to be equal to the

norm in the original (no-tilde) frame, i.e., to
xσ x σ = gσλ x σ x λ
Thus
gµν Λµ α Λν β x α x β = gσλ x σ x λ ; =⇒ gµν Λµ α Λν β x α x β = gαβ x α x β .
In the last step we have changed dummy indices to match the dummy
indices on the 4-vectors on both sides of the equation. This equality
has to be true for any 4-vector x. Therefore, the coefficients of x α x β
on the two sides have to be equal. Thus we have
gµν Λµ α Λν β = gαβ
This is our familiar expression ΛT gΛ = g (the definition of Lorentz

transformations), written in tensor/index notation instead of matrix
notation.

MP352notes IndexNotation A 01

Загружено:

Сведения о документе

Оригинальное название

Авторское право

Доступные форматы

Поделиться этим документом

Поделиться или встроить документ

Параметры публикации

Этот документ был вам полезен?

Это неприемлемый материал?

Авторское право:

Доступные форматы

MP352notes IndexNotation A 01

Загружено:

Авторское право:

Доступные форматы

p.

1 Tensor notation or index notation

2 Latin & Greek, Einstein notation

We have previously complained that the symbol X µ is often used to

For the Minkowski indices, we usually use lower-case Greek letters,

We are familiar with writing Lorentz transformations (from inertial

for spacetime coordinates. Here Λ is a 4 × 4 transformation matrix.

You recognize this as being just the definition of matrix multiplica-

Since any repeated index is a summation index and hence a dummy

The summation convention is widely used also in situations where no

might see this, e.g., in an advanced text on non-relativistic classical

I read somewhere that Einstein regarded the summation convention to be one

By definition, any 4-vector has to transform the same way under

3 Indices upstairs and downstairs — contravari-

3.1 Raising and lowering

variant 4-vector is obtained by contracting the contravariant index

The timelike component is unchanged by index lowering, while the

3.2 Inner products and norms

Aµ Bµ = Aµ Bµ = gµν Aµ Bν = gµν Aν Bµ = gµν Aµ Bν = gαβ A β Bα .

Hopefully, it should be obvious that Aµ Bµ = Aα Bα = A β B β =

Ãµ = g̃µν Ãν = gµν Ãν = gµν Λν σ Aσ = gµν Λν σ gσβ A β

Looks like the transformation matrix is gµν Λν σ gσβ . Remembering the

Now from the definition of the Lorentz transformation:

We have used ‘index lowering’ to define covariant vectors, and from

The transformation equation for covectors is sometimes written as

3.4 The metric tensor and its inverse, notation

3.5 Maybe think of covectors as row vectors?

The Lorentz transformation equation Ãµ =  0   0

index, serving as a row index and a column in-

In the row-vector picture, this can be visualized as

or equivalently, flipping signs of the spatial components first and

the row-vector picture is rather limited. The index notation is actu-

In quantum mechanics, this is a useful and widely accepted way to think

3.6 The 4-gradient and 4-divergence

We can generalize to the Minkowski world. The 4-derivative opera-

Thus, ∂µ φ transforms as a covector, justifying our use of subscripts

4 Tensors in non-relativistic mechanics (3-tensors)

Tij0 = ai0 b0j = Rik ak ∑ R jm bm = Rik ( ak bm )R jm = Rik Tkm (RT )mj

We have found the transformation rule T 0 = R T RT for the special

4.1 Examples of non-relativistic rank-2 tensors

which could also be written as a matrix-vector multiplication.

5 4-tensors and their transformations

gαµ T αβ = Tµ β , g βν T αβ = T α ν , gαµ g βν T αβ = Tµν .

and the totally covariant tensor Tρσ transforms as

You might find the Λ−1 ’s clumsy to write, especially to remember

Mαβµ . The outer product Aαβ Bµ ν is a rank-4 tensor with 3 contravari-

A tensor T µ1 µ2 ...µr β1 β2 ...β s of rank r + s, with r contravariant indices

6 Why is tensor notation useful?

Pαβ γ = Qαβ γ + Rαβ γ + Aαβ Bγ

holds in frame Σ. Of course, for the equation to make sense, each

implies element-by-element equality.) Do check that this is true in

P̃αβ γ = Q̃αβ γ + R̃αβ γ + Ãαβ B̃γ .

The reason is that, when transforming tensors, each uncontracted in-

7 The Lorentz transformation in index nota-

where we have used matrix notation: The LT’s Λ are represented

If the 4-coordinate x µ transforms to x̃ α under a LT, then its norm

Since the norm has to be invariant, we need this to be equal to the

gµν Λµ α Λν β x α x β = gσλ x σ x λ ; =⇒ gµν Λµ α Λν β x α x β = gαβ x α x β .

This is our familiar expression ΛT gΛ = g (the definition of Lorentz

Вам также может понравиться